Simulation of LUCC Scenarios and Analysis of the Driving Force of Carbon Stock Supply Changes in the North China Plain in the Context of Urbanization

Abstract

The North China Plain is the core region of China’s economic development, and exploring the impacts of its land use and cover change (LUCC) and different urbanization regional drivers on carbon stocks is conducive to promoting sustainable development and carbon balance within the region. In the study, the North China Plain was selected as the study area, and the Patch-Generating Land Use Simulation (PLUS) model and Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model were comprehensively applied to set up three land use policies, predict land use changes in 2030, and calculate carbon stock changes. Meanwhile, the Extreme Gradient Boosting (XGBoost) algorithm was used to analyze the degree of influence of different drivers on the supply of carbon stocks in different urbanization regions. Studies show that if the North China Plain prioritizes economic development, the area of farmland and forests will significantly decrease, leading to a substantial decline in carbon stocks. If ecological protection is the development focus, the reduction in farmland and forests will be less, and carbon stocks will remain relatively stable. If farmland protection is the development focus, the reduction in farmland will be minimal, but there will still be some impact on carbon stocks. The driving forces of carbon stock supply vary significantly across different regions. In underdeveloped regions, population density and vegetation cover have a greater impact on carbon stocks. In developing and urban–rural combined regions, vegetation cover and population migration have a greater impact on carbon stocks. In developed regions, the area of artificial land and gross domestic product (GDP) have a greater impact on carbon stocks. The study results provide scientific evidence for regional land use planning and policy formulation.

1. Introduction

Due to the intensifying effects of climate change and human activities [1], the impacts of LUCC on ecosystem services have become a growing concern, posing a significant challenge to the Earth’s ecosystem [2]. Between 1980 and 2020, the world’s population grew from 4.5 billion to 7.7 billion, with the population moving from rural areas to urban areas, leading to an increase in population density and urban area (in China, from 1981 to 2020, the area of urban built-up land increased from 6720 km² to 58,355.3 km² (obtained from the National Bureau of Statistics of China, http://www.stats.gov.cn/, Accessed 26 May 2024)). Especially in rapidly urbanizing areas, humans have changed land cover types by cutting down or transplanting trees [3], filling lakes [4], and discharging rivers [5], affecting the balance and long-term sustainability of regional ecosystems. Therefore, there is an urgent need to predict and assess future LUCC under different policies and management measures through scientific methods to guide rational LUCC planning and policy formulation. Land simulation scenario modeling, as an effective tool [6], can predict future LUCC trends under different scenario settings, assess the impacts of different policies, understand changes in ecosystem services, help balance the relationship between economic development and ecological protection, respond to potential environmental problems in advance, and provide diversified management options [7]. Therefore, research on the impact of LUCC on ecosystem services during urbanization is important for the development of scientific and rational LUCC planning and policies.

Currently, existing studies mainly focus on the impacts of urban expansion on the supply of ecosystem services [7,8], few studies consider detailed analyses of the driving factors behind the relationship of carbon stock changes in regions with different degrees of urbanization under different land modeling scenarios [9,10], and most studies have assessed the impact of human activities on carbon stock changes mainly based on changes in land attributes [11], without considering multiple factors such as population growth, economic development, and spatial changes in land [12,13,14]; in areas with high urbanization levels, rapid urban expansion and increased population density often lead to large-scale destruction of ecosystems with high carbon stocks such as forests and wetlands, which in turn exacerbates carbon emissions [6]. In contrast, areas with low levels of urbanization face the same problem of decreasing carbon stocks due to agricultural expansion and infrastructure development, although LUCC is slower [15]. In order to achieve sustainable urban development, the present urban ecosystem should be understood from a multidimensional perspective and the impact of human activities on carbon stock changes in areas with different urbanization levels should be measured. Only by considering these factors comprehensively, especially the need to focus on the differences between areas with different urbanization levels and exploring more adaptive sustainable development strategies, can the complex impacts of human activities on carbon stock changes be more accurately revealed, which in turn can provide a scientific basis for the formulation of more effective urban planning and management policies.

Various methods have been used at home and abroad to study the spatio-temporal variation patterns of LUCC. These methods include the Logistic-CA-Markov model [16], Least Squares Support Vector Machine–Class Activation Mapping (LSSVM-CA) model [17], Cellular Automata–Markov (CA-Markov) model [18], the Future Land Use Simulation (FLUS) model [19], and the Conversion of Land Use and its Effects at Small Regional Extent (CLUE-S) model [20]. However, these models are relatively weak in revealing the intrinsic drivers of LUCC and are unable to capture the patch evolution of multiple LUCC types, especially the complex changes in natural LUCC types. In recent years, most scholars [21,22] have begun to use the Patch-Generating Land Use Simulation (PLUS) model to capture the integrated probability of LUCC, which has the advantages of high simulation accuracy and rapid data processing, and can effectively simulate the complex evolution of multiple land types [21]. Since the reform and opening up, rapid urbanization in the North China Plain has led to significant changes in land use structure. Arable land and forest areas have gradually decreased, while urban construction land has rapidly increased, exacerbating the imbalance in carbon stocks and leading to pronounced changes in land use structure. This complex pattern of land use change makes the North China Plain an ideal region for studying the dynamic changes in carbon stocks and the impacts of land use policies. Additionally, as a critical economic and agricultural area, research findings from the North China Plain can provide scientific evidence for policy making and promote regional sustainable development. Therefore, to achieve urban sustainability and maintain regional carbon balance, it is essential to reveal the changes in carbon stocks under different land simulation scenarios and assess the extent of carbon stock changes in various urbanized areas.

This study focused on the North China Plain and employed the PLUS model to simulate LUCC data for 2015, 2018, and 2020. The accuracy of the simulated LUCC data was validated using the kappa coefficient, overall accuracy (OA) index, and cross-entropy loss function, enabling predictions for three LUCC scenarios in 2030. By integrating high-resolution remote sensing data and field measurements, the InVEST model’s carbon storage module was parameterized and regionally adapted to assess the long-term impacts of different land use change patterns, such as agricultural expansion, urban expansion, and forest restoration, on carbon storage. Additionally, building on existing research, various socio-economic factors, including population density, eco-economic development, and policy orientation, were comprehensively considered to analyze their driving effects on carbon storage. The XGBoost algorithm was used to investigate the impacts of different driving factors on carbon sequestration services across various urbanized regions. Finally, based on the multi-source data and analysis results obtained, practical regional carbon management strategies were proposed, emphasizing the importance of ecological protection while addressing economic development needs to achieve sustainable development in the North China Plain.

2. Materials and Methods

2.1. Study Area

The North China Plain (32°13′ N–40°26′ N, 112°17′ E–122°67′ E) is located in the lower reaches of the Yellow River, and is the second largest plain in China, accounting for 3.1% of China’s total land area, with a vast area and a low topography, with an elevation of mostly less than 50 m above sea level, and is a typical alluvial plain due to the deposition of a large amount of sediment carried by the Yellow River, the Huaihe River, the Haihe River, the Luanhe River, and so on, and in most places the deposits are as thick as seven or eight hundred meters. The lower reaches of the Yellow River run naturally through the middle of the North China Plain and are divided into two parts, north and south: the Yellow River and Huaihe River Plain in the south, and the Haihe River Plain in the north. Most of the North China Plain has a temperate monsoon climate and a subtropical monsoon climate. Winters are dry and cold, summers are hot and rainy, and springs are dry with little rain and strong evaporation. Drought is heavy in spring and flooding is common in summer. The average annual precipitation decreases from south to north with increasing latitude. The types of land use in the North China Plain mainly include arable land, grassland, forest and urban construction land. With the advancement of urbanization, the urbanization rate of the North China Plain has been increasing, and the urban population has been increasing. Urbanization has brought about a large consumption of land resources, a gradual decrease in the area of arable land, and changes in the agricultural structure. The study area is illustrated in Figure 1, where A–D represent different landscape types in various regions of the North China Plain. Specifically, A denotes the mountainous and hilly areas in Beijing, which include forested land and urban construction sites. B represents the arable land along the Yellow River in Shandong Province. C refers to arable land near a reservoir outlet in Shandong Province. D shows the arable land at the urban–rural fringe in Jiangsu Province.

2.2. Data Acquisition and Processing

All data used in this study are from data developed by authoritative institutions or organizations. All collected data were uniformly projected to the WGS1984 coordinate system in ArcGIS Pro 3.2, and in order to reduce the experimental error, the double-three convolution interpolation method was used to sample the raster data as 1 km. A specific description of the data is shown in

Table 1. Data sources.

Data Set	Platform
Earth Big Data Science Engineering Data Sharing Service System (2015 LUCC Data, 2018 LUCC Data, 2020 LUCC Data)	http://www.geodata.cn/ (Accessed 26 May 2024)
DEM	NASA/USGS/JPL-Caltech
Average annual precipitation	University of California Merced
Average annual temperature	NASA LP DAAC at the USGS EROS Center
Soil moisture (10 cm)	University of California Merced
Nighttime light	Earth Observation Group, Payne Institute for Public Policy, Colorado School of Mines
GDP	Data Centre for Resource and Environmental Sciences, Chinese Academy of Sciences (https://www.resdc.cn/, Accessed 26 May 2024)
Water area
Distance to town center
Road network Railway network	https://www.openstreetmap.org (Accessed 28 May 2024)
River systems
Population density	https://landscan.ornl.gov/ (Accessed 26 May 2024)
Vegetation cover	https://www.geodata.cn/ (Accessed 29 May 2024)
Man-made land area
A Prolonged Artificial Nighttime-light Dataset of China (1984–2020)	https://poles.tpdc.ac.cn (Accessed 22 May 2024)

.

The research flow of this study is shown in Figure 2. Firstly, the upper left part describes the process of simulating LUCC by the PLUS model, which contains multiple drivers, such as road traffic, natural climate, topography, and economic development, which have important impacts on land use change. At the same time, there are restricted areas, including nature reserves and wetland reserves, where land use is strictly limited. In terms of scenario setting, there are three different scenarios: the arable land protection scenario (CPS) prioritizes the protection of arable land, the ecological protection scenario (EPS) prioritizes the protection of the ecological environment, and the economic development scenario (EDS) prioritizes economic development. The upper right section shows the simulation quality test, which is validated using actual LUCC data (2015, 2018, and 2020). The assessment metrics include kappa and OA metrics to measure the accuracy of the simulation results and a cross-entropy loss function to optimize the model. After passing the model validation, the simulation of different LUCC scenarios for 2030 is carried out. The lower left part shows the LUCC under different modeling scenarios, and analyzes the LUCC under different policy scenarios (economic development, arable land protection, and ecological protection), and the change paths of land use types will be different under different scenarios. The lower middle part analyzes the carbon stock changes in the North China Plain provinces under different scenarios in combination with the carbon pool data, and determines the quality of carbon stock changes under different scenarios. The lower right part demonstrates the sensitivity analysis of urbanization to carbon stock under different LUCC scenarios, using nighttime lighting data for zoning the degree of urbanization and applying the XGBoost model for sensitivity analysis of carbon stock changes under different scenarios. The drivers and constraints are detailed in Figure 3.

2.3. Methodology

2.3.1. LUCC Scenario Setting

To explore the impacts of spatial and temporal changes of different LUCCs on carbon stocks, three land modeling scenarios were set up in this study. The first is the arable land protection scenario (CPS), which aims to prioritize food security and slow down the transfer of arable land to other land types, and the specific measures are mainly to limit the expansion of non-agricultural construction land and strengthen arable land protection [23,24]. Next, is the ecological protection scenario (EPS), which focuses on ecological development, increases the probability of conversion of arable land and grassland to forests and wetlands, and supports the transformation of arable land to ecological land use [25,26]. Finally, there is the economic development scenario (EDS), which is centered on economic development and allows large-scale conversion of land to construction land, promoting the urbanization and industrialization process and encouraging the conversion of land to construction land such as industrial parks and residential areas [27]. Each scenario was conducted in a 10-year cycle to simulate the LUCC in 2030. By comparing the LUCC under these three scenarios, the study analyzed their impacts on carbon stocks and found suitable land policies for LUCC development in the North China Plain.

2.3.2. Multi-Scenario Simulation Projections of LUCC Based on the PLUS Model

In the study, three LUCC scenarios of the North China Plain were simulated based on the characteristics and advantages of the PLUS model (PLUS v1.4 boxed) land use simulation. The Patch-Generating Land Use Simulation (PLUS) model [28] adopted the rule mining framework of the Land Expansion Analysis Strategy (LEAS) and a Cellular Automata model based on multi-type random seeds (CARS) [29]. In the PLUS model, a multi-type random patch seeding mechanism based on threshold descent is employed. Utilizing the Monte Carlo method, when the neighborhood effect parameter $k$ of LUCC reaches 0, the probability surface $O{P}_{i,k}^{1,t}$ for each LUCC type is as follows:

$$O{P}_{i,k}^{1,t}=\left\{\begin{array}{c}{P}_{i,k}^1\times D_k^{t}if{\Omega }_{i,k}^t=0andr<P_{i,k}^1\\ P_{i,k}^1\times {\Omega }_{i,k}^t\times D_k^{t}allothers\end{array}\right.$$

(1)

where $O{P}_{i,k}^{1,t}$ denotes the total probability of land use, $P_{i,k}^1$ denotes the probability of growth of type $k$ land use at unit $i$, and $u_k$ denotes the threshold for generating new patches for land use type $k$. $D_k^t$ is the adaptive driving factor, reflecting the effect of future demand on a given land use type. ${\Omega }_{i,k}^t$ denotes the unit neighborhood effect, representing the neighborhood composition element under land cover; $r$ is a random value from 0 to 1, the threshold for generating new patches of land use for land use type $k$, as decided by the model user.

$$If{\sum }_{k=1}^N\left|G_C^{t-1}\right|-{\sum }_{k=1}^N\left|G_C^t\right|<StepThen,d=d+1$$

(2)

$$\left\{\begin{array}{c}Change{P}_{i,c}^1>\tau andT{M}_{k,c}=1\\ Unchange{P}_{i,c}^1\le \tau orT{M}_{k,c}=0\end{array}\right.\tau ={\delta }^d\times r$$

(3)

In this context, $Step$ represents the time step required for simulating LUCC, $\delta$ denotes the decay coefficient, $r$ represents a normally distributed random value with a mean of 1, $d$ indicates the decay step size, and $G_C^t$ and $G_C^{t-1}$, respectively, denote the difference between current and future demands at iterations $t$ and $t-1$. $T{M}_{k,c}$ represents the transition matrix, determining the feasibility of transforming LUCC type $k$ into type $c$ within a certain range. Additionally, specific parameter settings for the PLUS model are as follows: in the LEAS module, the number of regression trees is set to 50; mTry is set to 13; and the acceptance rate is 0.01. In the CARS module, the default neighborhood size is 3, and patch generation is set to 0.9.

2.3.3. Carbon Storage Calculation Based on the InVEST Model

The study utilized the carbon storage module of the InVEST (Version 3.121) model to quantify carbon storage in the North China Plain [30]. For each type of LUCC, the carbon storage module estimated the carbon content of the four main carbon pools—aboveground carbon density, belowground carbon density, soil carbon density, and dead organic matter carbon density (measured in kg/m²)—using Equations 4 and 5. The carbon pool data were determined based on the land characteristics of the study area, in conjunction with national databases and research from other scholars [30,31].

$$C_i=C_{above}+C_{below}+C_{soil}+C_{dead}$$

(4)

$$C_{total}=\sum _{i=1}^n{A}_i{C}_i$$

(5)

The formula used to calculate carbon storage is defined as follows: $i$ represents the LUCC type; $C_i$ denotes the carbon density of the LUCC type; $C_{above}$, $C_{below}$, $C_{soil}$, and $C_{dead}$ represent the aboveground carbon density, belowground carbon density, soil carbon density, and dead organic matter carbon density of the land class (kg/m²); $C_{total}$ is the total carbon storage of the terrestrial ecosystem ($t$); $A_i$ is the area of LUCC type $i$ (km²); and $n$ is the number of LUCC types (in this study, $n$ represents 6 land use types).

2.4. Validation of Land Use Accuracy

In this study, the kappa metric, widely used in LUCC studies to assess model accuracy, was employed to evaluate the degree of conformity between predicted and observed data. The OA metric was also applied to quantify the ratio of correctly predicted LUCCs to the total number of observations. The credibility of the land prediction results of the PLUS simulation model was assessed using the cross-entropy loss function [32], which measures the variability between the predicted and true probability distributions by comparing them with observed data. Kappa and OA values exceeding 0.75 indicate that the LUCC simulation results are highly credible [33]. Meanwhile, when the cross-entropy loss function is small, it can better reflect the future LUCC scenario. The calculation formula is as follows:

$$Kappa={\displaystyle \frac{{OA}_O-{OA}_E}{\left(1-{OA}_E\right)}}$$

(6)

$${OA}_O={\displaystyle \frac{\left(\sum _{k=1}^n{OA}_{kk}\right)}{N}}$$

(7)

$$H\left(\mathrm{P},\mathrm{Q}\right)=-{\sum }_{i}P\left(i\right)logQ\left(i\right)$$

(8)

In this equation, ${OA}_{kk}$ denotes the number of samples correctly identified as belonging to LUCC category $k$, $n$ represents the number of LUCC categories, ${OA}_O$ indicates the overall accuracy of classification, and ${OA}_E$ reflects the likelihood of consistency between the simulated results and the current LUCC data. $P\left(i\right)$ is the $i$ th element of the true probability distribution $P$, $Q\left(i\right)$ is the ith element of the true probability distribution $Q$, and log denotes the natural logarithm.

2.5. XGBoost (Extreme Gradient Boosting) Model

XGBoost is a decision tree ensemble learning algorithm similar to random forests, implemented within a gradient boosting framework designed to improve efficiency, flexibility, and portability [34]. XGBoost examines the nonlinear relationship between the dependent variable and each feature while maintaining a high prediction accuracy [35]. In this study, the XGBoost algorithm was loaded in the Matlab software (MATLAB R2021b) to assess the impact of various drivers on carbon storage supply services. The objective function of XGBoost consists of a loss function and a regularization term [36], as shown in Equations (9) and (10), where Equation (10) is the regularization term.

$$Obj\left(\theta \right)=\sum _{i=1}^n\mathrm{L}\left(y_i,y_i^{^\left(t\right)}\right)+\sum _{k=1}^t\mathsf{\Omega }\left(f_k\right)$$

(9)

In Equation (9), $\mathrm{L}\left(y_i,y_i^{^\left(t\right)}\right)$ is the loss function, usually squared error or log loss, $\mathsf{\Omega }\left(f_k\right)$ is the regularization term to control model complexity, $t$ is the number of iterations, and $\theta$ is the model parameters.

$$\mathsf{\Omega }\left(f\right)=\gamma T+{\displaystyle \frac{1}{2}}\lambda {\sum }_{j=1}^T{\omega }_j^2$$

(10)

In Equation (10), $\gamma$ is the penalty coefficient controlling the number of leaf nodes, $T$ is the number of leaf nodes in the tree, $\lambda$ is the L2 regularization coefficient, and ${\omega }_j$ is the weight of the $j$th leaf node.

3. Results

3.1. LUCC under Different Modeling Scenarios

Based on the 2010 and 2020 LUCC data, different scenarios of LUCC in 2030 were simulated using the model. Prior to this, we simulated LUCC results for 2015, 2018, and 2020 based on the 2010 LUCC data and compared the simulated results with the actual LUCC data for 2015, 2018, and 2020. This comparative analysis was conducted to evaluate the feasibility of the PLUS model in simulating LUCC under different scenarios and the effectiveness of the simulation results. The study primarily used the kappa coefficient, OA coefficient, and cross-entropy loss function to quantify the discrepancy between the simulated results and actual results, as shown in

Table 2. Gap between LUCC modeling results and real results.

Simulated Year	Kappa (Kappa > 0.75)	OA (OA > 0.75)	Cross-Entropy Loss Function (0, +∞)
2015	0.9362	0.9584	0.3788
2018	0.9344	0.9566	0.4265
2020	0.9237	0.9459	0.4721

. The results for both the kappa metrics and OA metrics were >0.75, and the cross-entropy loss function values were also small, indicating that the simulated LUCC results of the model can represent future LUCC effectively.

From 2010 to 2020, LUCC in the North China Plain were dominated by arable land and forests (

Table 3. LUCC during 2010–2020.

LUCC Type	2010		2020		Area Change (km2)
	Area (km2)		Percentage (%)
Arable land	396,703	50.70%	386,330	49.47%	−10,373
Forest	228,442	29.25%	222,077	28.43%	−6365
Grassland	61,234.6	7.84%	59,741.9	7.65%	−1492.7
Waters	26,086.8	3.34%	26,195.1	3.35%	108.3
Urban	60,120	7.70%	76,265.8	9.77%	16,145.8
Bare land	616.41	0.08%	664.24	0.09%	47.83
Wetland	7812.04	1.00%	9740.68	1.25%	1928.64

), with grasslands, waters, urban construction land, bare land and wetlands accounting for a relatively small proportion, only about 20 percent of the total area. Compared with 2010, the land distribution has changed to different degrees, and there was a significant increase in urban construction land (16,145.8 km²) in 2020, especially on the eastern coast and around some inland cities. At the same time, the encroachment of urban construction on some arable land, especially around urban expansion areas, had reduced the area of arable land, while forest and grassland had also decreased by 6365 km² and 1492.7 km², respectively, along with the urbanization process, with the largest decrease being in forest land, which had decreased by 0.82%.

Based on the LUCC observed between 2010 and 2020, various LUCC scenarios for 2030 were simulated, as shown in Figure 4 and Figure 5. Compared to 2020, the degree of LUCC varies across different simulated scenarios. In the CPS depicted in Figure 4c, the LUCC patterns are similar to those in 2020, but urban construction land expands further, particularly in coastal and metropolitan areas. This indicates that under current policies, urbanization will continue to accelerate. Specifically, grassland, water bodies, and bare land decrease by 14,533,417.8 km² and 453.06 km², respectively, while arable land, forest, urban construction land, and wetland areas increase by 2415 km², 1461 km², 1379 km², and 69.49 km², respectively. In this scenario, water bodies mainly transition to arable land and wetlands, and grasslands mainly transition to urban built-up areas and forest land. Additionally, some wetlands are converted into arable land, forest, water bodies, and urban construction land. In the EPS, ecological protection measures are strengthened. The data indicate that, compared to Figure 4b, in the EPS, arable land and forest areas increase by 2315 km² and 989.12 km², respectively. The expansion of urban construction land is somewhat controlled, although urban areas continue to grow. Arable land is converted to urban construction land, and grasslands transition to forests. In the context of continuous urban expansion, implementing ecological protection measures can effectively mitigate and reduce the encroachment of urban expansion on the natural environment, thereby protecting arable land and forests.

In the 2030 EDS (Figure 4b), urban construction land significantly increases, particularly around economically developed areas and major cities. The differences in specific land types are notable: forest area increases by 871 km², water bodies decrease by 4869.9 km², bare land decreases by 376.34 km², and wetland area increases by 119.13 km². Land conversion primarily involves the transition of arable land and water bodies to urban construction land, followed by the transition of grassland and wetlands to urban areas. Grassland transitions into forest areas, and some water bodies are converted into wetlands. This land simulation scenario indicates that prioritizing economic development will lead to accelerated urban expansion, potentially resulting in greater encroachment and a reduction in arable land and other natural cover types. The enlarged regional maps in Figure 4(a1–e1) more clearly illustrate the specific impacts of urban expansion on arable land and other land types, as well as the land use adjustment strategies under different scenarios. In 2010, urban construction land was concentrated in core areas, arable land was widespread and continuous, and forests and water bodies remained in their natural state. By 2020, the scale of urban land had expanded significantly, arable land had decreased, and urbanization had increased ecological pressure. In the CPS (Figure 4(c1)), arable land area increases, and urban expansion is restricted, highlighting the priority given to agricultural protection. In the EDS (Figure 4(d1)), urban land expands significantly, with a priority on economic growth, placing greater pressure on ecological areas. In the EPS (Figure 4(e1)), forest and water body areas increase and urban expansion is restricted, emphasizing environmental protection and ecological restoration.

As shown in Figure 5, significant differences in land use type areas are observed due to the varying policy orientations, economic development needs, and ecological protection measures in the land simulation scenarios. This differences reflects the profound impact of different policies and priorities on land use. Overall, in all three scenarios, the arable land area shows a decreasing trend, with the most pronounced decrease in the EDS, indicating that policies prioritizing economic development may threaten arable land protection, leading to large areas being converted to urban construction land. In the EPS, the decrease in arable land is relatively small, indicating the effectiveness of ecological protection measures. Forest area also shows a decreasing trend in all three scenarios, with the greatest reduction in the EDS, reflecting a downward trend. In different scenarios, the decrease in forest area is minimized, and grassland and water body areas remain relatively stable. This indicates that ecological protection measures slow down the reduction in forest area and limit the conversion of grassland and water bodies. Urban construction land exhibits an increasing trend across all three scenarios, particularly in the EDS, indicating that policies prioritizing economic development accelerate the urbanization process, leading to extensive development of urban construction land. In contrast, growth in the city center is less pronounced, while the urban periphery experiences minimal growth. This illustrates that ecological protection measures can somewhat limit urban expansion and contribute to maintaining ecosystem stability.

3.2. Carbon Stock Changes in North China Plain Provinces under Different Modeling Scenarios

Using the InVEST model, carbon stocks were computed for different land simulation scenarios and for the years 2010 and 2020 in the North China Plain, subdividing the total carbon stock into various administrative regions, as shown in

Table 4. Changes in carbon stocks in different regions of the North China Plain (×10⁷).

Region\Year	2010	2020	CPS	EDS	EPS
	t/ha
Anhui	189.23	187.91	188.07	186.87	187.11
Beijing	27.17	26.75	26.81	26.65	26.84
Hebei	293.1	291.01	292.90	291.68	293.14
Henan	211.05	209.23	209.34	208.03	208.18
Jiangsu	104.6	102.88	103.11	101.19	101.64
Shandong	224.04	217.35	217.72	216.32	216.97
Tianjin	18.22	17.4	17.47	17.28	17.49

and Figure 6. The carbon stock in the North China Plain was 1.07 × 10⁹ t in 2010 and 1.05 × 10⁹ t in 2020, representing a decrease of 1.49 × 10⁸ t over the decade. Among the seven administrative regions, Shandong Province experienced the largest decrease in carbon stock, reducing by 6.69 × 10⁷ t, while Beijing municipality saw the smallest decrease, reducing by 0.42 × 10⁷ t. Compared to 2020, the carbon stock increased by 2.9 × 10⁷ t in the CPS, decreased by 1.16 × 10⁷ t in the EPS, and decreased by 4.51 × 10⁷ t in the EDS. Across the simulation scenarios, only the CPS showed an increase in carbon stock, while the carbon stocks in the other three scenarios exhibited a decreasing trend.

Combining Table 4 and Figure 6, carbon stocks in different regions of the North China Plain presented different change trends under different land simulation scenarios. From 2010 to 2020, Anhui Province saw a slight decrease in carbon stock by 1.32 × 10⁷ t, with reductions of 1.04 × 10⁷ t in the EDS and 0.8 × 10⁷ t in the EPS. In the CPS, carbon stock in Anhui Province increased by 0.16 × 10⁷ t, though this growth is minimal, amounting to only 0.09%. Beijing Municipality’s carbon stock decreased by 0.42 × 10⁷ t from 2010 to 2020, with reductions of 0.1 × 10⁷ t in the EDS, and increases of 0.06 × 10⁷ t and 0.09 × 10⁷ t in the CPS and EPS, respectively. And the carbon stock deficit areas gradually expanded, shifting towards the southwest and northeast directions. Hebei Province’s carbon stock is expected to decrease by 2.1 × 10⁷ t, with a decrease rate of 0.72%. Interestingly, Hebei Province’s carbon stock showed an increasing trend in three different land scenarios. Specifically, in the CPS it increased by 0.65%; in the EDS, it increased by 0.23%; and in the EPS, it increased by 0.74%. The carbon stock deficit areas are mainly distributed in the southern region, showing an expanding trend.

Additionally, Henan Province’s carbon stock decreased by 1.82 × 10⁷ t from 2010 to 2020. Except for an increase of 0.11 × 10⁷ t in the CPS, carbon stock decreased in all other land simulation scenarios. The carbon stock deficit areas were primarily distributed in the northeastern and central regions. In Jiangsu Province, the carbon stock saw the greatest decreases in the EDS and EPS, with reductions of 1.64% and 1.21%, respectively. Shandong Province experienced the largest decrease in carbon stock, with a reduction of 6.69 × 10⁷ t. Among them, the largest increase in carbon stock was observed in the CPS, with an increase of 0.37 × 10⁷ t. The carbon stock deficit areas were mainly located in the central region and southern coastal areas. In Tianjin Municipality, the carbon stock showed the most significant decline from 2010 to 2020, with a decrease of 4.56%. Except for increases of 0.46% in the CPS and 0.58% in the EPS, carbon stock showed a decreasing trend in the other two land simulation scenarios. The carbon stock deficit in the Binhai New Area was particularly pronounced in the EDS.

3.3. Sensitivity Analysis of Urbanization to Carbon Stocks under Different LUCC Scenarios

Different driving force parameters were selected for the sensitivity analysis of carbon stock changes in the land simulation scenarios in the North China Plain. Based on nighttime light images, the North China Plain was classified into regions with different degrees of urbanization to explore the extent of the influence of different drivers on the carbon stock under different land simulation scenarios in different classified regions with different degrees of urbanization. These urban areas include undeveloped areas (F1), urban–rural combined areas (F2), developing areas (F3), and developed areas (F4), and the different driving force parameters are vegetation cover, gross domestic product (GDP), population density, and manufactured land. According to the United Nations World Population Prospects 2019 report [13] and the International Energy Agency’s projected data, the population of the North China Plain is predicted to grow by 1% per year, the GDP is predicted to grow by 4.6% per year, and the vegetation cover by 4% [37], and the population density data, GDP data, and vegetation cover data are obtained for the year 2030 density data, and manufactured land data are obtained according to the simulated natural land development scenario [38], and the driving forces are shown in Figure 7. In underdeveloped areas, the carbon stock in the North China Plain was most sensitive to population density. In different LUCC scenarios, the impact of population density in the EPS and EDS was higher than in the CPS. The most significant factor affecting carbon stock in the simulation scenarios was vegetation cover, but its impact in the EPS and EDS remained higher than in the CPS. In all land simulation scenarios, only population density and vegetation cover had a lower impact in the CPS compared to the EPS and EDS. We believe that in the CPS, the vegetation is lush in underdeveloped areas and is far from human activity interference, weakening the influence of population density and vegetation cover. In peri-urban areas, the impact of population density and GDP in the CPS is higher than in the EDS and EPS, with population density having the greatest impact across the three land simulation scenarios, followed by vegetation cover.

In developing regions, although the degree of influence of driving forces was the same as in other areas, the effect of driving forces on different land simulation scenarios was completely opposite to that in underdeveloped areas. The impact of population density and vegetation cover in the CPS was higher than in the EPS and EDS, while the influence of the other two driving factors in the CPS was weaker than in the other land simulation scenarios. In developed regions, due to the higher area of manufactured land, this factor showed a greater degree of influence compared to the other three urban classification regions, although its impact was still lower than that of population density. We also found that the impact of GDP in the CPS was significantly higher than in the EDS and EPS. This strong influence was not observed in the driving force sensitivity analysis of the other three urban classification regions. Moreover, it was the only driving force in this level of urbanization that had a greater impact in the CPS than on the other two scenarios. We believe this relationship occurs because under the CPS urban expansion is restrained, leading to an increase in ecological land area. Therefore, changes in GDP inevitably lead to urban expansion and a reduction in ecological land, thus exerting a strong effect in the CPS.

4. Discussion

We applied the PLUS model to simulate land use changes under different scenarios and used the InVEST model to evaluate changes in carbon storage. Similar studies have been conducted by previous scholars [39]. For example, in contrast to the global study conducted by Zongyao Sha [31], which primarily relied on pre-existing global parameters, our research employs a locally optimized model based on an in-depth analysis of regional characteristics. This approach significantly improves the accuracy of carbon stock estimates for the North China Plain, given its complex and variable geographical conditions. Futhermore, it offers a novel methodological framework and valuable insights for addressing carbon stock estimation challenges in other regions with similarly complex terrains. In Ferng’s study [40] on the relationship between economic change, financial development, and ecological footprint, it was found that economic activities have a significant negative impact on the ecosystem. Wang and Zhang [41,42] found in their study of land use changes and their ecological effects in China that urbanization and industrialization processes have accelerated land use changes, especially the reduction in arable land and forests, thereby negatively impacting ecosystem services. Yang [43] explored the relationship between climate change, policy intervention, and land use change in their study and found that policy intervention has a significant impact on land use and ecosystem services. However, the research of previous scholars [43,44] has mainly focused on the macro level, analyzing the relationship between the ecosystem and human society from a broad perspective. In contrast, our study explores the impact of regional carbon storage changes in the context of urban expansion in China through specific scenario simulations and driving factor analysis. By quantifying the driving factors across different land simulation scenarios and urbanization levels, we provide more specific and actionable conclusions, further emphasizing the importance of policy intervention in land use management and the protection of ecosystem services.

LUCC is a major anthropogenic factor influencing carbon stocks in terrestrial ecosystems, and carbon stocks can, to some extent, reflect land use patterns [45,46]. Our study employs a multi-scenario land simulation approach, which differs from the study by Arfanuzzaman and Dahiya [47] in Southeast Asia. Their research focused on urban economic development and concluded that urbanization leads to irreversible carbon stock losses. In contrast, our study considers various future policy orientations and finds that, from 2010 to 2020, the LUCC patterns in the North China Plain showed an overall trend of increased urban construction land and decreased arable land and forest, influenced by natural, socioeconomic, and policy factors. During this decade, arable land and forest decreased by 17,738 square kilometers, while urban construction land increased by 16,145.8 square kilometers, which aligns with the findings of other scholars [48,49]. The primary reasons for this change were policy adjustments and population growth [50]. Furthermore, since 2010, the region has experienced rapid industrial restructuring and urbanization, further driving the demand for land for social production [51]. Although the conversion of unused land to construction land has somewhat increased the carbon stocks in the study area, encroachment on grasslands and arable lands due to urban construction [52] remains a primary cause of carbon stock loss in the North China Plain. To enhance the carbon sink capacity of the North China Plain and mitigate the impact of human activities, measures such as grazing bans and reseeding should be implemented to restore degraded grasslands. By formulating rational land use policies and constructing green infrastructure, it is possible to offset some of the carbon losses associated with urbanization.

Rapid urban expansion and changes in LUCC have adverse impacts on terrestrial ecosystem carbon sinks, leading to a decline in regional carbon sequestration capacity [53]. The swift expansion of urban construction land results in a reduction in forest, arable land, and grassland areas [54,55], particularly high-carbon-density forests and grasslands, significantly decreasing regional carbon storage [52]. In the three different LUCC simulation scenarios for 2030, total carbon storage shows a shrinking trend. Under the EDS, rapid urbanization negatively impacts the carbon supply of ecosystem services, with significant reductions in arable land and forest areas as large areas are converted to urban construction land. This aligns with policies introduced in recent years, such as the “Yangtze River Delta Urban Agglomeration Development Plan” and the “Zhejiang Province New Urbanization Promotion Outline.” These policies reflect the current trajectory of the North China Plain, emphasizing economic growth and urban expansion, which leads to encroachment into natural ecosystems and the weakening of ecosystem service functions. The EPS demonstrates strong ecological protection effects, mitigating the negative impact of LUCC on carbon storage to some extent, with relatively smaller reductions in arable land and forest areas, and a less pronounced decline in carbon storage. In the CPS, policies restrict the expansion of non-agricultural construction land and strengthen arable land protection, resulting in a smaller negative impact on carbon storage, which aligns with the views of other scholars on regional carbon storage changes [56,57]. Therefore, we believe that when formulating future LUCC planning and policies for the North China Plain, it is necessary to comprehensively consider the balance between ecological protection and economic development to achieve the region’s sustainable development goals. In particular, it is essential to strengthen ecological protection measures to reduce the negative impact of LUCC on ecosystem services, promoting the stability and sustainable development of the ecosystem service supply–demand relationship.

The study does not consider the supply–demand flows of ecosystem services between the North China Plain and adjacent external regions, nor does it consider the relationships between potential supply and demand areas and the study region [58,59]. In future research, we will consider cross-regional impacts to enhance the significance of our study. Additionally, we conduct a sensitivity analysis of driving factors using the XGBoost model. Unlike the study by Dong et al. [60], which focuses primarily on the linear impact of economic factors on carbon stocks, our research reveals the nonlinear interactions among multiple factors such as population density, economic development, and policy orientation, identifying urbanization as the most significant driver of carbon stock changes in the North China Plain (Figure 7). In underdeveloped regions, population density and vegetation cover significantly affect carbon stocks, as these areas are in the early stages of urbanization. The aggregation and migration of the population mainly lead to the expansion of construction land, affecting LUCC type conversion [15], which in turn impacts the overall regional carbon supply. In developing regions and urban–rural fringes, vegetation cover and population migration significantly impact carbon stocks, as these areas often experience rapid urbanization and rapid population growth [61], resulting in high LUCC intensity [61]. This puts tremendous pressure on forests and grasslands, forcing their conversion into construction land [54]. In developed areas, the area of construction land and GDP significantly impact carbon stocks. Higher levels of economic development in these regions increase the demand for housing, transportation, and public facilities, further promoting the expansion of construction land. This leads to increased regional carbon emissions and exacerbates the proportion of carbon emissions attributable to transportation and energy consumption. Therefore, we believe that future urban development needs to specifically consider the main driving factors and develop different LUCC plans and policies for regions at different stages of development to effectively mitigate the negative impacts of LUCC on carbon stocks.

The study employs both the InVEST and PLUS models to estimate carbon stocks, integrating geographic information systems (GIS), remote sensing technology, and ecological modeling to address the issue of inadequate prediction accuracy from using a single model in complex environments. This approach more accurately simulates the spatial distribution and dynamic changes in carbon stocks, thereby providing scientific support for regional carbon management. By designing three different land use scenarios—arable land protection (CPS), ecological protection (EPS), and economic development (EDS)—we systematically analyze the long-term impacts of land use changes under different policy orientations on carbon stocks, revealing the long-term impact pathways of land policies. The study demonstrates the potential increase or loss in carbon stocks under various policy interventions, offering policymakers scientifically-based decision support. Our research deeply explores the mechanisms by which multidimensional socioeconomic factors influence carbon stocks, overcoming the limitations of traditional single-factor analyses. We apply the XGBoost algorithm to analyze the complex impacts of factors such as population density, economic growth, and land use policies on regional carbon stocks, breaking through the constraints of traditional linear regression and effectively capturing the nonlinear relationships among variables. This reveals the driving mechanisms behind carbon stock changes in regions with varying levels of urbanization. Based on scenario simulation results and driving factor analysis, our study proposes a series of regional carbon management strategies aimed at optimizing land use allocation, strengthening green infrastructure, and promoting sustainable economic development to achieve a balance between ecological protection and economic development. These strategic recommendations have practical guidance significance not only for the North China Plain but also provide policy references for other similarly rapidly urbanizing regions.

5. Conclusions

This study focuses on the North China Plain and uses the LEAS and CARS models within the PLUS framework to simulate LUCC data for 2015, 2018, and 2020. The accuracy of the simulated data was verified using kappa metrics, overall accuracy (OA) metrics, and the cross-entropy loss function. Based on these results, three LUCC scenarios for 2030 (CPS, EPS, and EDS) were simulated. Additionally, the XGBoost algorithm was employed to quantify the impact of different driving factors (population density, economic activities, and LUCC types) on carbon storage supply services. The research results indicate the following.

If the future development of the North China Plain region focuses on economic development, the area of farmland and forests will decrease the most. A large amount of farmland and forests will be converted into urban construction land, leading to an increase in urban construction land of more than 20,000 km², which will inevitably result in a reduction in carbon stocks. If the region’s development focuses on ecological protection, due to the measures taken for ecological protection the reduction in farmland and forests will be less, the increase in urban construction land will be minimal, and the carbon stocks will remain relatively balanced. If the region’s development focuses on farmland protection, the reduction in farmland will be mitigated, but farmland will still be converted to urban construction land, which will still impact carbon stocks.

The impact of driving factors on carbon storage supply services varies significantly across different regions, with vegetation coverage and population density having the most notable impact on all scenarios. In underdeveloped areas, population density is the most sensitive factor affecting carbon storage changes, with its impact being significantly higher in the EPS and EDS compared to the CPS. In urban–rural junction areas, population density and GDP have a greater impact in the CPS than in the EPS and EDS. In developing areas, although the influence of driving factors is similar to that in other regions, population density and vegetation coverage have a higher impact in the CPS compared to the EPS and EDS, contrary to the pattern observed in underdeveloped areas. In developed regions, the area of manufactured land has a higher impact in the CPS, and changes in GDP have a much greater impact in the CPS than in the EPS and EDS.

Our study provides crucial data support for future urban LUCC and regional carbon emission reduction and carbon sequestration strategies, playing an important role in mitigating the negative impact of LUCC on ecosystem services. This study emphasizes the need to balance ecological protection and economic development when formulating LUCC plans and policies to achieve sustainable development goals. Strengthening ecological protection measures is particularly important to reduce the negative impact of LUCC on ecosystem services, thereby promoting the stability and sustainable development of ecosystem service supply and demand relationships.